Mapping plant populations with confidence

1

• Mapping plant populations with confidence !

The use of geostatistics in vegetation monitoring
Adrian R Yallop Jonathan I Thacker Cranfield
University
Simon James Joe Deimel Carl Bro
2
WHY?

• Measure responses to

Measure responses to

• large-scale environmental perturbation such as
  climatic change

local events changes in land management -
infrastructure developments - water extraction -
eutrophication - pollution etc.
These ideally require the ability to assess
fine-scale population response e.g. those that
might serve as warning of incipient larger
events
IN ADDITION
assessing conservation management or
bioremediation considerable amounts are spent -
but how effectively ?
we should be able learn from our
successes/failures towards Sutherland's
suggestions of 'evidence-based conservation'
Not to mention statutory obligations - national
and European legislation - CBD etc. most
require data on inventory change etc to be
reported
3
Existing protocols for used for mapping and
monitoring REMOTE SENSING - HABITAT
CLASSIFICATION METHODS

• Remote sensing continues to be suggested as a
  'viable' option for monitoring - however

Classification accuracies typically 80-85 -
whatever the methods - always have been
! consider also

• these are average accuracies - some classes hard
  to get wrong 99-100 those we might want maybe
  down to 40-50!

• reported academic results frequently represent
  months of work on a single image by researchers -
  no connection with a deployable repeatable method
  at all

• broad classes - defined by what remsen can do -
  rather than what ecologists might need may work
  for landscape scale e.g LCM but not finer

There have been few tests against actual
conservation criteria - especially for monitoring
Irrespective of future improvements in spatial,
spectral or temporal resolution there is little
chance such methods will ever achieve the
taxonomic resolution required for field or
site-scale conservation monitoring
these are only ever likely to be obtained by
combination of field survey and analysis precise
enough to allow valid statistical temporal and
spatial comparison
4
Difficulties associated with existing protocols
for used for mapping and monitoring FIELD
SURVEY - COMMUNITY CLASSIFICATION METHODS

• boundary definition in NVC subjective
  (arbitrary?)
• by eye - hand drawn
• no confidence can be placed on location - same
  problem with changes over time

• quadrat location biased
• hence no statistics between time periods possible

• heavily dependent on surveyor commitment and
  competence
• level of sub community extraction

In summary changes in vegetation, or surveyor ? -
no history of NVC validation - comparison of
results - demonstration of the method for
monitoring etc.

• Other issues with community extraction methods
• community extraction often does not utilise the
  species we may be interested in
• e.g. rare or indicator species

These are NOT criticisms of NVC
but rather the application of the method for
processes for which it was not designed and is
poorly suited
5
Difficulties associated with existing mapping and
monitoring protocols FIELD SURVEY - POPULATION
ASSAY

• quadrat sampling not spatially enabled
• hence no information about distribution within
  the sample area

• statistics invariably use data aggregation
• i.e. means of quadrat data

• this is specifically designed to overcome spatial
  variance in the samples
• yet this is the very data we want !

populations vary in space as well as time
responses to management, perturbations etc.
likely to be manifest as spatial change within a
site before significant overall change is
apparent - gives earlier warning e.g.
population response across a site might be
indicative of changing hydrology - sea level rise
- eutrophication etc.
there are existing methods - fixed repeat visit
quadrats reveal spatial information BUT
relocation slow - limits sample numbers - i.e.
the process of working out where to put quadrat
takes as long as looking at the vegetation
we need a reliable/robust vegetation mapping
method with accuracies that allow assay and
monitoring which is ABOVE ALL simple enough for
routine deployment and removes 'fixed-quadrat'
overhead
6
Suggestions

• 2 main technological changes may provide help -
  the accuracy and availability of
• Sub-metre GPS data-loggers
• GIS and geostatistical software

these are not necessarily NEW - but rather now
cheap and simple enough for routine deployment
we have been looking at combining these two
technologies to develop a new way of visualising
plant population distribution and change,
especially for monitoring
we have NOT developed this from an academic
viewpoint rather that it should be sufficiently
robust and efficient in implementation to allow
relatively routine application
especially it should NOT require any more
investment in software, skills or equipment than
could be considered reasonable i.e everything
required should be readily available
7
Progress ?
Well obviously we have some !

• Field Survey

Subjective mapping replaced by a protocol of
unbiased random sampling
easiest method is simply walking site in paced
straight lines (transects) quadrats at
predetermined distance - each transect same
distance apart as each sample - in effect the
site is visited as grid - typically between
10-20m
at each sample location
quadrat position is recorded to sub-metre
accuracy using DGPS - ALL species present within
quadrat are recorded as presence/absence -
datalogged
overcomes problems of phenology reduces time -
no need to estimate abundance - frequency - check
etc.

• No subjective estimates of location or abundance
  - lines on paper maps or 'fag-packets' -
• just 'is the plant' here

• Quick - accurate - unbiased - REPEATABLE

8
Sampling Stiffkey Marsh 2002
9
Mapping Analysis

• Interpolation is routinely used in other
  disciplines to create surfaces from point data
• many types - but to map field-scale variation
  - require local interpolators

• some - like IDW - use arbitrary weighting of
  observed values in generating predictions

for monitoring we require the mapping process to
be repeatable - requires freedom from individual
choices - we want to map vegetation NOT arbitrary
decisions

• Kriging uses spatial relationships within the
  data to generate prediction weighting

e.g. is the plant homogenous across the sampled
area or is it patchy

• This is determined by plotting relationship
  between pairs of observations at differing
  distances - a semivariogram - prediction
  weighting is determined by line fitting

e.g. the nature of the distribution observed
determines the nature of the prediction used

• It IS a complex topic BUT

in reality application easy using Geostatistical
analyst in Arc - important in making this a
general tool
10
Mapping Analysis - continued
We produce maps using

• Kriging interpolation - Indicator kriging (for
  1/0 data)

used to produce a map of Probability (Pmap) that
a value exceeds 0 i.e. in this case the
probability of a plant occurring at a given point
Model parameterisation

• A crucial part of our method is that is it should
  be insensitive to quadrat location
• i.e. future surveys need not visit the same
  sample sites - just repeat the basic method -
  keeps it quick

seek to smooth consequences of individual samples
- kriging relatively insensitive to lag distance
greater than 2D - increasing neighbour number in
point estimation also smoothes surface
We choose to use lag 2D and neighbours 20 to
a great extent this choice can be seen as
arbitrary BUT should be fixed between surveys
11
Example of Pmap generation. Aster tripolium at
Stiffkey marsh 2002
12
Example of Pmap generation. Ulex europaea at
Stiffkey marsh 2002
13
Change Analysis

• At one level this is simple !

Pmap t1 - Pmap t0 creating a Dmap i.e. a
map of differences in probability of a species
occurring at a given point

• This of will course contain all the small
  differences arising from chance within the
  process
• in particular those resulting from sampling
  position

• Crucial to the method has been finding ways of
  dealing with this - development of a 95
  confidence interval for mapped change using a
  random map of same freq

Dmaps can if required be 'thresholded' according
to the distribution of changes identified i.e.
we assign a confidence interval to changes
In effect we produce a map of changes at Plt0.05
or Plt0.01 (i.e. maps showing areas were we can
be be 95 or 99 certain change has occurred)
95 or 99 Cmaps

• What does all this mean ?

They say a picture is worth a thousand words !!
14
(No Transcript)
15
(No Transcript)
16
(No Transcript)
17
Other benefits

• Field protocol NOT tied to analysis method - data
  can be re-analysed by different methods in future
  - including, if required, community extraction

compared to existing methods which cannot be
deconstructed to reveal species distribution
information or allow different analyses

• Every sample location known - can be revisited
  

either to monitor individual plants OR check ID
- AND the surveyor !

• Rare species like orchids, can be individually
  mapped

• Species data are stored directly within a GIS

database queries - visualisation/analyses within
full GIS environment including orthocorrected
imagery - OS mapping etc etc etc
18
Festuca pratensis Cricklade 1998 - 2004
19
Festuca pratensis Cricklade 1998 - 2004
20
Eco-hydrology extraction impacts on SSSI
integrity

• Collection of spatially enabled point data for
  all species allows mapping of ecological
  indicators such Ellenberg indices

This is now being used to assess the impacts of
hydrological extraction on SSSI integrity for AWS
for a number of sites

• Two way in which Ellenberg values can be
  generated as part of this process
• 1 directly from all the species Pmaps from the
  sites
• 2 by interpolation of Ellenberg values from each
  sample.

Both of these are open to the same Dmaps and
Cmap thresholding to show areas of change in
ecological indicators over time
Again a picture is worth a thousand words !!
21
Example of visualisation of mapping of Ellenberg
values Middle Harling Fen
22
Summary

• New technologies can give us the tools to answer
  the questions we, as ecologists, want answering

• in particular the ability to really monitor
  changes in plant populations

can provide repeatable unbiased and above all
defendable assessments of plant distribution and
change

• It is time we moved into the realms of best
  practice and gathered the data required to make
  conservation ecology and EIA the evidence-based
  science it ought to be

• In essence it seems we are still using many of
  the tools and concepts of early last century

23
(No Transcript)